Model fit utilities

This repository contains useful functions for evaluating model fit. It's currently fairly limited, but will (probably) be expanded in the future.

Installation

To install this package, clone the repository and run pip install . -e from the root directory. It can then be used as a regular Python package, e.g. from model_fit_tools.plotting import plot_recovery.

To install other dependencies, run pip install -r requirements.txt.

Documentation

Documentation is available here.

Examples

To plot parameter recovery, use the plot_recovery function. For example:

from model_fit_tools.plotting import plot_recovery

# Estimate parameters using your favourite method
# This can be either point estimates or posterior samples
parameter_estimates = model_fit(data)

# Plot parameter recovery
plot_recovery(
    true_values,  # The true parameter values used to simulate data
    parameter_estimates,  # The estimated parameter values
    ['alpha', 'beta', 'gamma']  # The names of the parameters
)

To plot a probability-probability (PP) plot (to assess posterior calibration) use the plot_pp function. For example:

from model_fit_tools.plotting import plot_pp

# Estimate parameters using your favourite method
# Must be in the form of posterior samples
parameter_estimates = model_fit(data)

# Plot PP plot
plot_pp(
    true_values,  # The true parameter values used to simulate data
    parameter_estimates,  # The estimated parameter values
    ['alpha', 'beta', 'gamma']  # The names of the parameters
)